Practical Multi-Task Learning for Rare Conversions in Ad Tech

2025-07-27

AI Summary

To address low prediction accuracy for rare conversion events (occurring <1% of the time) in online advertising, this paper proposes a frequency-aware multi-task learning (MTL) framework. It groups conversion types by occurrence frequency and employs a shared bottom architecture with frequency-specific task towers, enabling joint modeling of both frequent and rare conversions. Historical statistical priors are incorporated to guide conversion-type grouping, thereby enhancing model sensitivity to sparse signals. Offline evaluation shows a 0.69% AUC improvement; online deployment reduces cost-per-action by 2%, and the method is now fully operational in production. The core contribution lies in a frequency-driven tower design within a unified MTL framework, which explicitly captures the distributional characteristics of rare events—significantly boosting their prediction performance while maintaining scalability and practicality.

Technology Category

• Machine Learning: Online Learning & Bandits
• Reasoning under Uncertainty: Relational Probabilistic Models
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We present a Multi-Task Learning (MTL) approach for improving predictions for rare (e.g., <1%) conversion events in online advertising. The conversions are classified into "rare" or "frequent" types based on historical statistics. The model learns shared representations across all signals while specializing through separate task towers for each type. The approach was tested and fully deployed to production, demonstrating consistent improvements in both offline (0.69% AUC lift) and online KPI performance metric (2% Cost per Action reduction).

Problem

• Improving predictions for rare conversion events in online advertising
• Classifying conversions into rare or frequent types based on history
• Learning shared representations while specializing via task towers

Innovation

• Multi-Task Learning for rare conversions
• Shared representations with task towers
• Deployed with offline and online improvements